From Vector Databases to Knowledge Engines: The Next Layer of AI (46 min)

← Back to week overview

• Release date: 2026-05-05
• Listen on Spotify: Open episode
• Episode description:

  Peter Levine speaks with Ash Ashutosh, CEO of Pinecone, about the launch of Nexus and the shift from vector databases to knowledge engines. As agents become the primary users of software, they discuss why traditional retrieval systems break down and how AI systems need to evolve to support machine-to-machine interactions. The conversation explores how agents currently spend most of their time retrieving and reasoning over data, why that approach is inefficient, and how moving reasoning closer to the data can dramatically improve performance, accuracy, and cost. Ash also explains how Pinecone is rethinking the stack for agentic applications, introducing new abstractions, query languages, and developer workflows.   Resources: Follow Ash Ashutosh on X: https://x.com/ashashutosh Follow Peter Levine on LinkedIn: https://www.linkedin.com/in/peter-levine-681386172/   Check out everything a16z is doing with artificial intelligence here, including articles, projects, and more podcasts. Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Summary

• 🚀 Shift to Agent Users: Pinecone observed a massive user shift from humans to AI agents 8-9 months ago, revealing that 85% of agent work is inefficient knowledge retrieval on human-centric systems.
• 🔧 Nexus Knowledge Engine: Nexus evolves vector databases into context-compiling knowledge engines, creating task-specific artifacts for structured, cited responses that boost accuracy over 90% and cut times dramatically.
• 📊 Efficiency Gains: Internal tests showed token usage drop 90% (40k to 2k), response times from minutes to <500ms, and task completion from <50% to >90%, offloading LLM burdens to data proximity.
• 🗣️ NoQL Protocol: The new NoQL query language standardizes agent interactions with intent, timing, and governance parameters, aiming to become the industry standard like SQL for databases.
• 🌐 Future Ecosystem: With marketplaces, open standards, and revised pricing, Nexus enables a Cambrian explosion of trusted, vertical agent apps, simplifying enterprise AI from demo to production.

Insights

• What if AI agents could bypass the inefficiencies of human-designed data systems to complete tasks faster and more accurately?
  • Time: 0:12 – 6:19
  • Answer: Pinecone’s Nexus shifts retrieval from brute-force queries suited for humans to context-compiled knowledge engines, reducing token usage by up to 90% and boosting task completion from under 50% to over 90%. This matters because it addresses the core bottleneck in agentic AI, where 85% of work is knowledge retrieval, enabling scalable enterprise applications.
• How can compiling context directly at the data source transform AI reasoning from reactive to proactive?
  • Time: 11:21 – 14:17
  • Answer: Nexus moves reasoning closer to data curation, creating task-specific artifacts that provide structured, cited responses, unlike LLMs that reason post-retrieval on potentially incomplete info. This enhances accuracy, explainability, and trust in enterprise settings by attributing answers to sources.
• Could offloading AI retrieval to specialized engines spark a new era of cost-efficient agentic applications?
• Why might a new query language like NoQL become the SQL of AI agents?
• In what ways will trusted knowledge engines redefine enterprise AI adoption?
• How might marketplaces and standardized stacks accelerate the build-out of agentic AI ecosystems?
• What pricing evolution could make AI infrastructure accessible at agent scale?

Want the full treatment? Download expanded insights—timestamps, complete answers, and course materials from AI Weekly Insights.